Cellxgenedp

cellxgenedp

Workflows

Standard Workflow

The cellxgene data portal (https://cellxgene.cziscience.com/) provides a graphical user interface to collections of single-cell sequence data processed in standard ways to 'count matrix' summaries. The cellxgenedp package provides an alternative, R-based inteface, allowind data discovery, viewing, and downloading.

library(cellxgenedp)
library(dplyr)

# 1. Initialize database
cellxgene_db <- db()

# 2. Discover collections, datasets, and authors
cols <- collections(cellxgene_db)
dsets <- datasets(cellxgene_db)
auths <- authors()

# 3. Query datasets by specific author
author_datasets <- left_join(
  authors(),
  datasets(),
  by = "collection_id",
  relationship = "many-to-many"
)

# 4. Filter by disease using facets
covid_datasets <- author_datasets |>
  filter(facets_filter(disease, "label", "COVID-19"))

Inputs/Outputs: Takes online metadata from the CELLxGENE data portal retrieved via db(), and outputs filtered tibbles of collections, datasets, and authors matching specific search criteria.

When to Use

• To programmatically discover, filter, and retrieve single-cell RNA-seq datasets from the CELLxGENE data portal.
• To query datasets by specific authors or consortia using authors() and datasets().
• To identify datasets matching specific ontological terms (e.g., disease, cell type, tissue) using facets() and facets_filter().
• To launch an interactive Shiny interface for data discovery and download using cxg().

When NOT to Use

• For downstream single-cell analysis like clustering, dimensionality reduction, or differential expression (use scran, scater, or Seurat instead).
• For analyzing private, non-public single-cell datasets not hosted on the CELLxGENE portal.

Data Requirements

• No local input files are required to start; the package retrieves metadata dynamically from the CELLxGENE portal via db().
• Downloaded datasets are typically in H5AD format, which can be loaded into R using zellkonverter.

Key Parameters

• relationship ("many-to-many"): Parameter used in left_join when merging authors and datasets.
• facet ("disease"): Column name passed to facets() to retrieve ontology terms and counts.
• label ("COVID-19"): Target ontology term label used in facets_filter().

Best Practices

• Use db() to fetch the latest database state from the CELLxGENE portal at the start of your session.
• Use distinct(authors()) to deduplicate author records before performing downstream joins, as some collections contain duplicate author entries.
• Use facets(db(), "disease") or other facet fields to inspect available ontology terms and their exact labels before filtering.

Common Pitfalls

• Joining authors and datasets directly without handling many-to-many relationships, leading to warnings. Fix: Specify relationship = "many-to-many" in left_join().
• Assuming a single disease per dataset. Fix: Remember that the disease column is a list-of-lists because a dataset can contain multiple diseases; use facets_filter() or tidyr::unnest_longer() to handle it.

Alternatives

• scRNAseq for accessing curated public single-cell datasets from Bioconductor.
• HCAData for accessing Human Cell Atlas data.

Citations

• Morgan M (2024). cellxgenedp: Discover and retrieve single-cell data from the CELLxGENE portal. R package version 1.16.0.

References

• Homepage: bioconductor.org/packages/cellxgenedp
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/cellxgenedp/inst/doc/Discovery_and_retrieval.html